finally got around to reading this eccv20 best paper award winner work on optical flow that has interesting ideas: multi-scale 4D correlation volumes, learned-upsampling (using convex weights of lower res pixels), and iterative refinement of flow.
arxiv.org/abs/2003.12039
arxiv.org/abs/2003.12039
correlation volumes
upsampling module
better optical flow helps in better depth estimation
https://twitter.com/ankurhandos/status/1220755115256057856
RAFT iteratively refines optical flow with GRU. Differentiable primal-dual is another interesting approach to iteratively refine solutions
arxiv.org/abs/1607.07988
arxiv.org/abs/1607.07988
optical flow
- teaches euler-lagrange, primal-dual optimisation methods
- shows how to add regularisation to ill-posed problems.
- often a doorway to understanding inverse problems in computer vision e.g. depth estimation, denoising, inpainting, superres.
- teaches euler-lagrange, primal-dual optimisation methods
- shows how to add regularisation to ill-posed problems.
- often a doorway to understanding inverse problems in computer vision e.g. depth estimation, denoising, inpainting, superres.
@ducha_aiki is this any different to transposed convolution/upconvolution/deconvolution?
• • •
Missing some Tweet in this thread? You can try to
force a refresh
